1EXECVE(2) Linux Programmer's Manual EXECVE(2)
2
6 execve - execute program
7
9 #include <unistd.h>
10 int execve(const char *pathname, char *const argv[],
12 char *const envp[]);
13
15 execve() executes the program referred to by pathname. This causes the
16 program that is currently being run by the calling process to be re‐
17 placed with a new program, with newly initialized stack, heap, and
18 (initialized and uninitialized) data segments.
19 pathname must be either a binary executable, or a script starting with
21 a line of the form:
22 #!interpreter [optional-arg]
24 For details of the latter case, see "Interpreter scripts" below.
26 argv is an array of pointers to strings passed to the new program as
28 its command-line arguments. By convention, the first of these strings
29 (i.e., argv[0]) should contain the filename associated with the file
30 being executed. The argv array must be terminated by a NULL pointer.
31 (Thus, in the new program, argv[argc] will be NULL.)
32 envp is an array of pointers to strings, conventionally of the form
34 key=value, which are passed as the environment of the new program. The
35 envp array must be terminated by a NULL pointer.
36 The argument vector and environment can be accessed by the new pro‐
38 gram's main function, when it is defined as:
39 int main(int argc, char *argv[], char *envp[])
41 Note, however, that the use of a third argument to the main function is
43 not specified in POSIX.1; according to POSIX.1, the environment should
44 be accessed via the external variable environ(7).
45 execve() does not return on success, and the text, initialized data,
47 uninitialized data (bss), and stack of the calling process are over‐
48 written according to the contents of the newly loaded program.
49 If the current program is being ptraced, a SIGTRAP signal is sent to it
51 after a successful execve().
52 If the set-user-ID bit is set on the program file referred to by path‐
54 name, then the effective user ID of the calling process is changed to
55 that of the owner of the program file. Similarly, if the set-group-ID
56 bit is set on the program file, then the effective group ID of the
57 calling process is set to the group of the program file.
58 The aforementioned transformations of the effective IDs are not per‐
60 formed (i.e., the set-user-ID and set-group-ID bits are ignored) if any
61 of the following is true:
62 * the no_new_privs attribute is set for the calling thread (see
64 prctl(2));
65 * the underlying filesystem is mounted nosuid (the MS_NOSUID flag for
67 mount(2)); or
68 * the calling process is being ptraced.
70 The capabilities of the program file (see capabilities(7)) are also ig‐
72 nored if any of the above are true.
73 The effective user ID of the process is copied to the saved set-user-
75 ID; similarly, the effective group ID is copied to the saved set-group-
76 ID. This copying takes place after any effective ID changes that occur
77 because of the set-user-ID and set-group-ID mode bits.
78 The process's real UID and real GID, as well as its supplementary group
80 IDs, are unchanged by a call to execve().
81 If the executable is an a.out dynamically linked binary executable con‐
83 taining shared-library stubs, the Linux dynamic linker ld.so(8) is
84 called at the start of execution to bring needed shared objects into
85 memory and link the executable with them.
86 If the executable is a dynamically linked ELF executable, the inter‐
88 preter named in the PT_INTERP segment is used to load the needed shared
89 objects. This interpreter is typically /lib/ld-linux.so.2 for binaries
90 linked with glibc (see ld-linux.so(8)).
91 Effect on process attributes
93 All process attributes are preserved during an execve(), except the
94 following:
95 * The dispositions of any signals that are being caught are reset to
97 the default (signal(7)).
98 * Any alternate signal stack is not preserved (sigaltstack(2)).
100 * Memory mappings are not preserved (mmap(2)).
102 * Attached System V shared memory segments are detached (shmat(2)).
104 * POSIX shared memory regions are unmapped (shm_open(3)).
106 * Open POSIX message queue descriptors are closed (mq_overview(7)).
108 * Any open POSIX named semaphores are closed (sem_overview(7)).
110 * POSIX timers are not preserved (timer_create(2)).
112 * Any open directory streams are closed (opendir(3)).
114 * Memory locks are not preserved (mlock(2), mlockall(2)).
116 * Exit handlers are not preserved (atexit(3), on_exit(3)).
118 * The floating-point environment is reset to the default (see
120 fenv(3)).
121 The process attributes in the preceding list are all specified in
123 POSIX.1. The following Linux-specific process attributes are also not
124 preserved during an execve():
125 * The process's "dumpable" attribute is set to the value 1, unless a
127 set-user-ID program, a set-group-ID program, or a program with capa‐
128 bilities is being executed, in which case the dumpable flag may in‐
129 stead be reset to the value in /proc/sys/fs/suid_dumpable, in the
130 circumstances described under PR_SET_DUMPABLE in prctl(2). Note
131 that changes to the "dumpable" attribute may cause ownership of
132 files in the process's /proc/[pid] directory to change to root:root,
133 as described in proc(5).
134 * The prctl(2) PR_SET_KEEPCAPS flag is cleared.
136 * (Since Linux 2.4.36 / 2.6.23) If a set-user-ID or set-group-ID pro‐
138 gram is being executed, then the parent death signal set by prctl(2)
139 PR_SET_PDEATHSIG flag is cleared.
140 * The process name, as set by prctl(2) PR_SET_NAME (and displayed by
142 ps -o comm), is reset to the name of the new executable file.
143 * The SECBIT_KEEP_CAPS securebits flag is cleared. See capabili‐
145 ties(7).
146 * The termination signal is reset to SIGCHLD (see clone(2)).
148 * The file descriptor table is unshared, undoing the effect of the
150 CLONE_FILES flag of clone(2).
151 Note the following further points:
153 * All threads other than the calling thread are destroyed during an
155 execve(). Mutexes, condition variables, and other pthreads objects
156 are not preserved.
157 * The equivalent of setlocale(LC_ALL, "C") is executed at program
159 start-up.
160 * POSIX.1 specifies that the dispositions of any signals that are ig‐
162 nored or set to the default are left unchanged. POSIX.1 specifies
163 one exception: if SIGCHLD is being ignored, then an implementation
164 may leave the disposition unchanged or reset it to the default;
165 Linux does the former.
166 * Any outstanding asynchronous I/O operations are canceled
168 (aio_read(3), aio_write(3)).
169 * For the handling of capabilities during execve(), see capabili‐
171 ties(7).
172 * By default, file descriptors remain open across an execve(). File
174 descriptors that are marked close-on-exec are closed; see the de‐
175 scription of FD_CLOEXEC in fcntl(2). (If a file descriptor is
176 closed, this will cause the release of all record locks obtained on
177 the underlying file by this process. See fcntl(2) for details.)
178 POSIX.1 says that if file descriptors 0, 1, and 2 would otherwise be
179 closed after a successful execve(), and the process would gain priv‐
180 ilege because the set-user-ID or set-group-ID mode bit was set on
181 the executed file, then the system may open an unspecified file for
182 each of these file descriptors. As a general principle, no portable
183 program, whether privileged or not, can assume that these three file
184 descriptors will remain closed across an execve().
185 Interpreter scripts
187 An interpreter script is a text file that has execute permission en‐
188 abled and whose first line is of the form:
189 #!interpreter [optional-arg]
191 The interpreter must be a valid pathname for an executable file.
193 If the pathname argument of execve() specifies an interpreter script,
195 then interpreter will be invoked with the following arguments:
196 interpreter [optional-arg] pathname arg...
198 where pathname is the pathname of the file specified as the first argu‐
200 ment of execve(), and arg... is the series of words pointed to by the
201 argv argument of execve(), starting at argv[1]. Note that there is no
202 way to get the argv[0] that was passed to the execve() call.
203 For portable use, optional-arg should either be absent, or be specified
205 as a single word (i.e., it should not contain white space); see NOTES
206 below.
207 Since Linux 2.6.28, the kernel permits the interpreter of a script to
209 itself be a script. This permission is recursive, up to a limit of
210 four recursions, so that the interpreter may be a script which is in‐
211 terpreted by a script, and so on.
212 Limits on size of arguments and environment
214 Most UNIX implementations impose some limit on the total size of the
215 command-line argument (argv) and environment (envp) strings that may be
216 passed to a new program. POSIX.1 allows an implementation to advertise
217 this limit using the ARG_MAX constant (either defined in <limits.h> or
218 available at run time using the call sysconf(_SC_ARG_MAX)).
219 On Linux prior to kernel 2.6.23, the memory used to store the environ‐
221 ment and argument strings was limited to 32 pages (defined by the ker‐
222 nel constant MAX_ARG_PAGES). On architectures with a 4-kB page size,
223 this yields a maximum size of 128 kB.
224 On kernel 2.6.23 and later, most architectures support a size limit de‐
226 rived from the soft RLIMIT_STACK resource limit (see getrlimit(2)) that
227 is in force at the time of the execve() call. (Architectures with no
228 memory management unit are excepted: they maintain the limit that was
229 in effect before kernel 2.6.23.) This change allows programs to have a
230 much larger argument and/or environment list. For these architectures,
231 the total size is limited to 1/4 of the allowed stack size. (Imposing
232 the 1/4-limit ensures that the new program always has some stack
233 space.) Additionally, the total size is limited to 3/4 of the value of
234 the kernel constant _STK_LIM (8 MiB). Since Linux 2.6.25, the kernel
235 also places a floor of 32 pages on this size limit, so that, even when
236 RLIMIT_STACK is set very low, applications are guaranteed to have at
237 least as much argument and environment space as was provided by Linux
238 2.6.22 and earlier. (This guarantee was not provided in Linux 2.6.23
239 and 2.6.24.) Additionally, the limit per string is 32 pages (the ker‐
240 nel constant MAX_ARG_STRLEN), and the maximum number of strings is
241 0x7FFFFFFF.
242
244 On success, execve() does not return, on error -1 is returned, and er‐
245 rno is set to indicate the error.
246
248 E2BIG The total number of bytes in the environment (envp) and argument
249 list (argv) is too large.
250 EACCES Search permission is denied on a component of the path prefix of
252 pathname or the name of a script interpreter. (See also
253 path_resolution(7).)
254 EACCES The file or a script interpreter is not a regular file.
256 EACCES Execute permission is denied for the file or a script or ELF in‐
258 terpreter.
259 EACCES The filesystem is mounted noexec.
261 EAGAIN (since Linux 3.1)
263 Having changed its real UID using one of the set*uid() calls,
264 the caller was—and is now still—above its RLIMIT_NPROC resource
265 limit (see setrlimit(2)). For a more detailed explanation of
266 this error, see NOTES.
267 EFAULT pathname or one of the pointers in the vectors argv or envp
269 points outside your accessible address space.
270 EINVAL An ELF executable had more than one PT_INTERP segment (i.e.,
272 tried to name more than one interpreter).
273 EIO An I/O error occurred.
275 EISDIR An ELF interpreter was a directory.
277 ELIBBAD
279 An ELF interpreter was not in a recognized format.
280 ELOOP Too many symbolic links were encountered in resolving pathname
282 or the name of a script or ELF interpreter.
283 ELOOP The maximum recursion limit was reached during recursive script
285 interpretation (see "Interpreter scripts", above). Before Linux
286 3.8, the error produced for this case was ENOEXEC.
287 EMFILE The per-process limit on the number of open file descriptors has
289 been reached.
290 ENAMETOOLONG
292 pathname is too long.
293 ENFILE The system-wide limit on the total number of open files has been
295 reached.
296 ENOENT The file pathname or a script or ELF interpreter does not exist.
298 ENOEXEC
300 An executable is not in a recognized format, is for the wrong
301 architecture, or has some other format error that means it can‐
302 not be executed.
303 ENOMEM Insufficient kernel memory was available.
305 ENOTDIR
307 A component of the path prefix of pathname or a script or ELF
308 interpreter is not a directory.
309 EPERM The filesystem is mounted nosuid, the user is not the superuser,
311 and the file has the set-user-ID or set-group-ID bit set.
312 EPERM The process is being traced, the user is not the superuser and
314 the file has the set-user-ID or set-group-ID bit set.
315 EPERM A "capability-dumb" applications would not obtain the full set
317 of permitted capabilities granted by the executable file. See
318 capabilities(7).
319 ETXTBSY
321 The specified executable was open for writing by one or more
322 processes.
323
325 POSIX.1-2001, POSIX.1-2008, SVr4, 4.3BSD. POSIX does not document the
326 #! behavior, but it exists (with some variations) on other UNIX sys‐
327 tems.
328
330 One sometimes sees execve() (and the related functions described in
331 exec(3)) described as "executing a new process" (or similar). This is
332 a highly misleading description: there is no new process; many at‐
333 tributes of the calling process remain unchanged (in particular, its
334 PID). All that execve() does is arrange for an existing process (the
335 calling process) to execute a new program.
336 Set-user-ID and set-group-ID processes can not be ptrace(2)d.
338 The result of mounting a filesystem nosuid varies across Linux kernel
340 versions: some will refuse execution of set-user-ID and set-group-ID
341 executables when this would give the user powers they did not have al‐
342 ready (and return EPERM), some will just ignore the set-user-ID and
343 set-group-ID bits and exec() successfully.
344 On Linux, argv and envp can be specified as NULL. In both cases, this
346 has the same effect as specifying the argument as a pointer to a list
347 containing a single null pointer. Do not take advantage of this non‐
348 standard and nonportable misfeature! On many other UNIX systems, spec‐
349 ifying argv as NULL will result in an error (EFAULT). Some other UNIX
350 systems treat the envp==NULL case the same as Linux.
351 POSIX.1 says that values returned by sysconf(3) should be invariant
353 over the lifetime of a process. However, since Linux 2.6.23, if the
354 RLIMIT_STACK resource limit changes, then the value reported by
355 _SC_ARG_MAX will also change, to reflect the fact that the limit on
356 space for holding command-line arguments and environment variables has
357 changed.
358 In most cases where execve() fails, control returns to the original ex‐
360 ecutable image, and the caller of execve() can then handle the error.
361 However, in (rare) cases (typically caused by resource exhaustion),
362 failure may occur past the point of no return: the original executable
363 image has been torn down, but the new image could not be completely
364 built. In such cases, the kernel kills the process with a SIGSEGV
365 (SIGKILL until Linux 3.17) signal.
366 Interpreter scripts
368 The kernel imposes a maximum length on the text that follows the "#!"
369 characters at the start of a script; characters beyond the limit are
370 ignored. Before Linux 5.1, the limit is 127 characters. Since Linux
371 5.1, the limit is 255 characters.
372 The semantics of the optional-arg argument of an interpreter script
374 vary across implementations. On Linux, the entire string following the
375 interpreter name is passed as a single argument to the interpreter, and
376 this string can include white space. However, behavior differs on some
377 other systems. Some systems use the first white space to terminate op‐
378 tional-arg. On some systems, an interpreter script can have multiple
379 arguments, and white spaces in optional-arg are used to delimit the ar‐
380 guments.
381 Linux (like most other modern UNIX systems) ignores the set-user-ID and
383 set-group-ID bits on scripts.
384 execve() and EAGAIN
386 A more detailed explanation of the EAGAIN error that can occur (since
387 Linux 3.1) when calling execve() is as follows.
388 The EAGAIN error can occur when a preceding call to setuid(2), se‐
390 treuid(2), or setresuid(2) caused the real user ID of the process to
391 change, and that change caused the process to exceed its RLIMIT_NPROC
392 resource limit (i.e., the number of processes belonging to the new real
393 UID exceeds the resource limit). From Linux 2.6.0 to 3.0, this caused
394 the set*uid() call to fail. (Prior to 2.6, the resource limit was not
395 imposed on processes that changed their user IDs.)
396 Since Linux 3.1, the scenario just described no longer causes the
398 set*uid() call to fail, because it too often led to security holes
399 where buggy applications didn't check the return status and assumed
400 that—if the caller had root privileges—the call would always succeed.
401 Instead, the set*uid() calls now successfully change the real UID, but
402 the kernel sets an internal flag, named PF_NPROC_EXCEEDED, to note that
403 the RLIMIT_NPROC resource limit has been exceeded. If the PF_NPROC_EX‐
404 CEEDED flag is set and the resource limit is still exceeded at the time
405 of a subsequent execve() call, that call fails with the error EAGAIN.
406 This kernel logic ensures that the RLIMIT_NPROC resource limit is still
407 enforced for the common privileged daemon workflow—namely, fork(2) +
408 set*uid() + execve().
409 If the resource limit was not still exceeded at the time of the ex‐
411 ecve() call (because other processes belonging to this real UID termi‐
412 nated between the set*uid() call and the execve() call), then the ex‐
413 ecve() call succeeds and the kernel clears the PF_NPROC_EXCEEDED
414 process flag. The flag is also cleared if a subsequent call to fork(2)
415 by this process succeeds.
416 Historical
418 With UNIX V6, the argument list of an exec() call was ended by 0, while
419 the argument list of main was ended by -1. Thus, this argument list
420 was not directly usable in a further exec() call. Since UNIX V7, both
421 are NULL.
422
424 The following program is designed to be execed by the second program
425 below. It just echoes its command-line arguments, one per line.
426 /* myecho.c */
428 #include <stdio.h>
430 #include <stdlib.h>
431 int
433 main(int argc, char *argv[])
434 {
435 for (int j = 0; j < argc; j++)
436 printf("argv[%d]: %s\n", j, argv[j]);
437 exit(EXIT_SUCCESS);
439 }
440 This program can be used to exec the program named in its command-line
442 argument:
443 /* execve.c */
445 #include <stdio.h>
447 #include <stdlib.h>
448 #include <unistd.h>
449 int
451 main(int argc, char *argv[])
452 {
453 char *newargv[] = { NULL, "hello", "world", NULL };
454 char *newenviron[] = { NULL };
455 if (argc != 2) {
457 fprintf(stderr, "Usage: %s <file-to-exec>\n", argv[0]);
458 exit(EXIT_FAILURE);
459 }
460 newargv[0] = argv[1];
462 execve(argv[1], newargv, newenviron);
464 perror("execve"); /* execve() returns only on error */
465 exit(EXIT_FAILURE);
466 }
467 We can use the second program to exec the first as follows:
469 $ cc myecho.c -o myecho
471 $ cc execve.c -o execve
472 $ ./execve ./myecho
473 argv[0]: ./myecho
474 argv[1]: hello
475 argv[2]: world
476 We can also use these programs to demonstrate the use of a script in‐
478 terpreter. To do this we create a script whose "interpreter" is our
479 myecho program:
480 $ cat > script
482 #!./myecho script-arg
483 ^D
484 $ chmod +x script
485 We can then use our program to exec the script:
487 $ ./execve ./script
489 argv[0]: ./myecho
490 argv[1]: script-arg
491 argv[2]: ./script
492 argv[3]: hello
493 argv[4]: world
494
496 chmod(2), execveat(2), fork(2), get_robust_list(2), ptrace(2), exec(3),
497 fexecve(3), getauxval(3), getopt(3), system(3), capabilities(7), cre‐
498 dentials(7), environ(7), path_resolution(7), ld.so(8)
499
501 This page is part of release 5.13 of the Linux man-pages project. A
502 description of the project, information about reporting bugs, and the
503 latest version of this page, can be found at
504 https://www.kernel.org/doc/man-pages/.
505Linux 2021-08-27 EXECVE(2)